Expected goals (xG) measures the quality of a team's or player's chances: each shot is given the probability that a shot like it is scored, and xG is the sum of those probabilities. FIFA describes xG as reflecting "the average probability of scoring a goal with an individual attempt on goal".
A side that finishes a match on 1.60 xG made chances worth 1.6 goals on average, whether it scored none or four. In betting, xG is one input for estimating how many goals each side will score, which a model then turns into prices.
How expected goals is worked out
Every attempt gets a value from 0 to 1. Providers build their own models, but the usual inputs include the distance and angle to goal, the body part, how the chance was created and whether it was a penalty. A team's match xG adds up its shots, and because the models differ, two providers can give one match different figures.
The total hides how it was made. Two sides on 1.60 xG, with each shot treated as independent (illustrative):
| Side | Chances | xG | Chance of not scoring |
|---|---|---|---|
| Team A | 16 shots at 0.10 | 16 x 0.10 = 1.60 | 0.90^16 = 18.5% |
| Team B | 2 at 0.55 and 5 at 0.10 | 1.10 + 0.50 = 1.60 | 0.45^2 x 0.90^5 = 12.0% |
A few big chances make a goal more likely than many poor ones. A simple Poisson model on 1.60 goals puts the blank at e^-1.6 = 20.2% (e is the constant 2.718), higher than either shot mix gives.
Expected goals vs goals over a short run
Goals scatter widely around xG over a handful of matches. Say a side creates 9.6 xG across six games, 1.6 a game, and scores five (illustrative). If 9.6 were exactly right, a Poisson model puts five or fewer goals at e^-9.6 x (1 + 9.6 + 46.08 + 147.46 + 353.89 + 679.48) = 0.0000677 x 1,237.51 = 8.4%, about one run in 12.
So a short run rarely tells you whether a side is finishing badly or is unlucky. Goals are what bets settle on; xG says what the chances were worth. The longer the sample, the more weight a gap between them can bear.
Where xG misleads
- Chances that never become shots, such as a cross that just evades a striker, carry no xG.
- A side that leads early may sit deep, so the shots it allows later can make its defence look worse than it is.
- One penalty adds a large value, so a side's xG can rise on refereeing decisions rather than on how it plays.
- Red cards, injuries and changed line-ups shift a side's chances in ways its past xG cannot show.
Using xG in a pricing model, carefully
- Use a long sample, not the last five games, and pull it toward the league average when it is short.
- Look at penalties separately rather than letting a few decisions drive the rate.
- Adjust for home ground, team news and the opponent.
- Turn each side's expected goals into scoreline chances with a Poisson model, then into fair prices.
Say the steps give an expected total of 2.30 goals when the true figure is 2.15 (illustrative):
| Expected total | Two goals or fewer | Over 2.5 goals | Fair over price | An over price of $2.60 |
|---|---|---|---|---|
| 2.30 | e^-2.3 x (1 + 2.3 + 2.645) = 59.6% | 40.4% | 1 / 0.40396 = 2.4755, about $2.48 | 2.60 / 2.4755 - 1 = 5.0% above fair |
| 2.15 | e^-2.15 x (1 + 2.15 + 2.3113) = 63.6% | 36.4% | 1 / 0.36385 = 2.7484, about $2.75 | 2.60 / 2.7484 - 1 = -5.4%, below fair |
A 0.15-goal error in the input turns an apparent edge into a shortfall.
Soccer betting strategy builds the full method, both teams to score prices that market from the same inputs, and soccer betting covers the markets a model can price.
B337 is not a ratings tool: it has no model and does not pick bets with one. The Terminal puts bookmakers' soccer head to head, line and total prices next to each other. From Terminal View, its EV overlays compare each price with a fair price taken from a global market reference with the margin removed: an estimate, not a forecast, and something to check your own numbers against. A free account opens a limited view of the Terminal with live odds.
For free and confidential support call 1800 858 858 or visit gamblinghelponline.org.au.
Risk: Betting involves risk. A model's fair price carries real error, a bet above fair can lose and often does, and there is no guarantee of profit. See responsible gambling for limits and support.